Radial Attention: O(nlogn) Attention for Long Video Generation with 2-4× Speedup

3 lmxyy 1 7/7/2025, 8:31:10 PM hanlab.mit.edu ↗

Comments (1)

lmxyy · 14m ago
Introduce Radial Attention — a static sparse attention mechanism with O(nlogn) complexity for long video generation! Here are some key features: * Plug-and-play: works with pretrained models like Wan, HunyuanVideo, Mochi * Speeds up both training&inference by 2–4×, without quality loss * Compatible with pre-trained LoRAs. When applied to 8-step FusionX LoRA, Radial Attention further delivers a 1.6× speedup